An open API service indexing awesome lists of open source software.

https://github.com/jhrcook/statistical-rethinking

Notes on Statistical Rethinking: A Bayesian Course with Examples in R and Stan
https://github.com/jhrcook/statistical-rethinking

bayesian bayesian-statistics r rstats statistical-rethinking statistics

Last synced: about 2 months ago
JSON representation

Notes on Statistical Rethinking: A Bayesian Course with Examples in R and Stan

Awesome Lists containing this project

README

        

---
output: github_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

# Notes on *Statistical Rethinking: A Bayesian Course with Examples in R and Stan*

The first two chapters of the 2nd edition of this book are available as a [PDF](statisticalrethinking2_chapters1and2.pdf) for free.
The rest of the book I used the first edition as it is available through my schools library.

## Introduction

This book is an introduction to Bayesian statistics, focussing on providing an applicable education in R.
This repository contains my notes; these notes are not a thorough recapitulation of the book, but instead acting as a combination of a reference and a playground for myself.

These notes have been compiled into a simple static website for ease of reference and searching.

[![notes website](https://img.shields.io/badge/Notes_on-Statistical_Rethinking-bf344b.svg?style=flat)](https://statistical-rethinking-notes.netlify.app)

## Setting-up

This course includes an R packages called 'rethinking']().
It can be installed as follows.

```r
# Dependencies
install.packages(c("coda", "mvtnorm", "devtools", "loo", "dagitty"))

# Course package
devtools::install_github("rmcelreath/rethinking")
```

## Chapter Notes

[Chapter 1. The Golem of Prague](ch1_the-golem-of-prague.md)

[Chapter 2. Small Worlds and Large Worlds](ch2_small-worlds-and-large-worlds.md)

[Chapter 3. Sampling the Imaginary](ch3_sampling-the-imaginary.md)

[Chapter 4. Linear Models](ch4_linear-models.md)

[Chapter 5. Multivariate Linear Models](ch5_multivariate-linear-models.md)

[Chapter 6. Overfitting, Regularization, and Information Criteria](ch6_overfitting-regularization-and-information-criteria.md)

[Chapter 7. Interactions](ch7_interactions.md)

[Chapter 8. Markov Chain Monte Carlo](ch8_markov-chain-monte-carlo.md)

[Chapter 9. Big Entropy and the Generalized Linear Model](ch9_big-entropy-and-the-generalized-linear-model.md)

[Chapter 10. Counting and Classification](ch10_counting-and-classification.md)

[Chapter 11. Monsters and Mixtures](ch11_monsters-and-mixtures.md)

[Chapter 12. Multilevel Models](ch12_multilevel-models.md)

[Chapter 13. Adventures in Covariance](ch13_adventures-in-covariance.md)

[Chapter 14. Missing Data and Other Opportunities](ch14_missing-data-and-other-opportunities.md)

---

## Other R packages for Bayes

Below are some of the common packages for using Baysian statistics in R.
The purpose of this collection is to serve as a reference for future work.
As I work through their vignettes (and possible other associated tutorals/articles), the markdown files will be linked below.

**['rstanarm'](https://mc-stan.org/rstanarm/index.html)**

- [my notes](bayes-packages/rstanarm.md)

**['brms'](https://paul-buerkner.github.io/brms/)**

**['ggmcmc'](https://cran.r-project.org/web/packages/ggmcmc/index.html)**

**['tidybayes'](http://mjskay.github.io/tidybayes)**

- [my notes](bayes-packages/tidybayes_vignettes.md)

**['bayetestR'](https://easystats.github.io/bayestestR/)** (part of the 'easystats' suite of packages)